Understanding consumer online shopping behaviour from the perspective of transaction costs

5.5 PRELIMINARY RESULTS 
The proposed research model (refer Figure 3.1) depicts a total of 16 constructs to be 
measured within the model. Among these 16 constructs, 14 constructs (except online buying 
frequency and online purchase behaviour) were measured by multiple items. The ensuring 
preliminary analysis follows the procedure as set out in Figure 5.1 to individually examine 
the psychometric properties of each of these 14 constructs’ scales.
5.5.1 Preliminary Analysis –Internet Access Availability 
Four items (IAA1 to IAA4) were used to measure Internet access availability
 
construct. 
These were subjected to the two-step preliminary data analysis process shown in Figure 5.1, 
the results of which appear in Table 5.2. Evaluation of the correlation matrix through the 
KMO and Bartlett’s Test results in a high KMO statistic (.787) and a significant probability 
level (p< .001) for the Bartlett’s test. These results indicate that sufficient correlations were 
found within the correlation matrix for factor analysis to proceed. In addition, bivariate 
correlations were inspected and all coefficients fell within the acceptable range for factor 
analysis of .30 and .90. EFA was then conducted which produced a single factor structure 
with strong item loading ranging from .78 to .88 and the variance explained was 70.29%.

---

216
Cronbach’s alpha of .86 was computed indicating good reliability of the scale. At this point, 
as all the items met the criteria of Step One, they were retained for CFA analysis in Step Two. 
The four items (retained from Step One) measuring Internet access availability were 
subjected to CFA in Step Two (refer Figure 5.1) of the preliminary analysis. The model fit 
indices indicated that this measurement model did not fit quite well to the data (
χ
2
/df = 29.08 > 
5, RMSEA = .171 < .08, and RMR =.066 > .05). The poor fit of the model was then revised 
by investigating modification indices and the standard residuals (Hair
 et al.
2006). 
Specifically, indicators which were related to larger reductions of chi-square or problematic 
standard residuals (2.5 as cut-off) (Anderson and Gerbing 1988) were identified and 
eliminated one by one. After removing item IAA4, an acceptable model was achieved. All 
overall goodness-of-fit statistics were within acceptable ranges. For example, 
χ
2
/df (.023) was 
less than 5, GFI (0.986), TLI (0.992) and CFI (0.983) were above .90, RMSEA (.002) was 
less than .08 and RMR (.002) was less than .05. As displayed in Table 5.2, the standardized 
factor loadings of the remaining three items ranged from .68 to .85. All exceeded the 
minimum standardized loading .50 with two of them above the ideal criterion of .70 (Hair
 et 
al.
2006). The AVE was .63, exceeding the recommended criterion of .50 (Hair
 et al.
2006). 
The CR of this construct was .834, exceeding the threshold of .70 (Nunnally 1978). As a 
result, items IAA1, IAA2 and IAA3 were considered to constitute a reliable and valid 
measurement scale for Internet access availability. These three items were retained for 
constructing the overall measurement model. 

---

217